CONCLUSIONS

Three main conclusions can be drawn from this paper:

• A variety of tools can be used to support the implementation of multiuser applications. Some of them such as shared object systems extend programming language abstractions, others such as distributed RPC systems extend communication abstractions, while still others such as shared window systems, GroupKit, Rendezvous, and multiuser Suite extend user-interface abstractions such as windows, widgets, views, and active variables, respectively.

• Each of these tools has both important advantages and disadvantages. More important, each one of them offers some benefits that are not offered by other tools and thus supports some class of applications better than the others. Database systems offer efficient querying of large data structures and are thus suitable for traditional applications such as airline reservation systems that support independent multiuser interaction. Distributed RPC and constraint systems provide a high-level mechanism for keeping arbitrary relationships among distributed data and are thus suitable for applications such as the Mercury multiuser module editor, which allows users to make consistent changes to related, distributed modules. Message servers provide flexible mechanisms allow clients for different users to be unaware of each other. GroupKit supports extensible multiuser interfaces to applications developed using an existing single-user toolkit. Rendezvous provides high-level support for sharing of both semantic data and their screen representation while making no assumptions about how the screen representations are generated. Multiuser Suite provides high-level and efficient support for a variety of sharing, concurrency control, communication, and access control schemes, and allows users and applications to change these schemes dynamically.
• The approach of using a single-user tool as a basis for a multiuser tool must be taken to support collaborative interaction requiring sharing of interaction state that is not stored in client data structures. It is for this reason that almost all of the current work in multiuser tools is being carried out by the user-interface community.

In this paper, we have considered only a subset of the collaboration functions to compare the automation, flexibility, and performance of multiuser tools. It is necessary to also consider other important collaboration functions such as support for multimedia [Man91], undo/redo [Cho92], and navigation [Jef92] in multiuser applications. It would be also useful to consider other criteria for comparing these tools such as how easy it is to use the applications supported by them.

One of the important conclusions of this paper is that each of these tools has important advantages and drawbacks. We see two future approaches for solving this problem. One approach is to make these tools ``open'', that is, allow a single client to use the services of multiple tools, thereby allowing the application to use different multiuser tools to share different parts of its data. A second, more difficult and challenging, approach is to build an integrated system that offers the benefits of all of these tools. By describing the main concepts behind the design of these tools and pointing out the similarities and differences among the approaches implemented by them, this paper has taken a first step towards this goal.